JPS6057410A - Learning control method - Google Patents

Abstract

PURPOSE:To perform the accurate positioning with use of an inexpensive computer by varying the dynamic delay time to repeat the preliminary reproductions and calculating the integration value of a series of deviations to define the dynamic delay time with which the integration value is minimum as the delay time of a reproduction mode. CONSTITUTION:A deviation is calculated at the time (t+ntc) between a reproduction locus Y0 and the teaching value R0, and this deviation E(t+ntc) is added to the teaching value R0 which is earlier by the time ntc to obtain the target value R1(t). With this target value, the 1st preliminary reproduction is carried out. In this case, a preliminary reproduction locus Y1 is obtained. A deviation between the locus Y1 and the value R0 is detected between time ti and tj. The absolute values of those epsilon1. In the same way, the integration value epsilon2 is obtained. These values epsilon1 and epsilon2 are compared with each other. Hereafter the preliminary reproductions are repeated to obtain integration values epsilon3, epsilon4.... Then a delay time with which a series of these integration values are minimum is detected. This dynamic delay time is used to following operations of reproduction.

Description

[Detailed Description of the Invention] The present invention relates to a learning control method for objects that are repeatedly controlled, such as playbank robots, etc., and is particularly unaffected by noise, etc., and achieves 100 accuracy using inexpensive equipment] This invention relates to a learning control method that aims at -.

Generally, when performing positioning control of an object that involves repeated control as described above, first perform a teaching operation to make the object memorize the position data (teaching value) of the target work trajectory, and then perform the teaching operation according to the teaching value. While performing regenerative operation, a learning control method is adopted in which a deviation between the taught value and the driving trajectory is detected, and this deviation is added to the taught value and used as a target value for the next regenerative operation. ing. However, in this case, simply adding the deviation to the taught value or the current target value cannot eliminate the control deviation due to dynamic delay, and if the object has multiple degrees of freedom, energization, The dynamic delay time differs for each degree of freedom, and the amount of deviation also changes from time to time for each degree of freedom. Therefore, simply adding the same amount of deviation to all degrees of freedom will result in poor synthesis of the tip of the controlled object. It is not possible to reduce the deviation below a certain level, and it is necessary to make detailed corrections for each degree of freedom.

Therefore, we measure the dynamic delay time moment by moment for each degree of freedom.
The present inventor has developed a microscopic learning control method in which the deviation is added to the taught value or the target value earlier by this changing dynamic delay time and reproduced. This method allows extremely fine-grained positioning control, which greatly improves control accuracy, but it requires a storage device with a large storage capacity, and not only can relatively low-cost microcomputers not be used. However, since the control is too microscopic, it is strongly influenced by noise, etc., and has the drawback that the accuracy does not reach the expected culm.

Therefore, the object of the present invention is to determine macroscopically whether or not the measured delay time is appropriate, that is, whether or not the deviation is minimized, thereby reducing the influence of noise. The purpose of this method is to provide a learning control method that performs accurate positioning control with a small amount of memory without being affected by Perform regenerative operation according to the teaching instructions, measure the deviation between the taught value and the re-trajectory, and then perform the next regenerative operation! ! iIJ
The present invention provides a learning control method in which the deviation is added to the teaching value or the current target trajectory as early as the time corresponding to the target delay time, and 1: driving is performed again.

Next, with reference to the attached drawings, an embodiment of the present invention will be explained in order to provide an understanding of the present invention.

Here, FIG. 1 is a block diagram showing a signal transmission system of a control circuit that can be used to implement the present invention, and FIG. 2 is a conceptual diagram showing the contents of a temporary storage device RAM used in the control circuit.
FIG. 3 does not show an example of the processing procedure of the learning control method which is an embodiment of the present invention] U-charts 1 and 4 are graphs for explaining the present invention in detail.

In the following embodiments, a learning control method for a play bank type robot made using a hydraulic cylinder will be explained as an example, but the learning control method in the present invention can be applied not only to a robot using such a hydraulic cylinder as a drive source. Needless to say, the present invention is applicable to robots using various electric motors as drive sources, and various other white gJs machines.

FIG. 1 shows only the control system for one degree of freedom of a hydraulic cylinder-driven robot as shown in I-, and the other degrees of freedom are omitted because they are the same.
Further, although the case where a microcomputer is used as the control circuit is shown as an example, it is of course possible to replace this with various other well-known control elements.

In Figure 1, the computer 1 has a central processing unit CPU2 that performs arithmetic processing, a ROM 3 that stores the processing program of this CPU2, and a
It consists of a temporary storage device RAM4 (see Figure 2) that temporarily stores various coefficients and variables necessary for performing depth calculation processing, but it also includes a human interface circuit that is naturally required. , output interface circuit, and other conversion means are well-known elements and are therefore omitted here.

The CPU 2 has a D/A converter 5. A servo valve 8 that operates a hydraulic cylinder 9 is connected through a comparator 6 and an amblifier 7, and a robot robot movable element 10 such as an arm, a swivel table 9, and a wrist mechanism is connected to the hydraulic cylinder 9. The amount of expansion and contraction of the hydraulic cylinder 9 and the amount of rotation of the movable element 10 are detected by the detector 11, converted into electrical signals, and then fed back to the comparator 6 and CI) U2.

fν・Required target (i
rtR & all are stored in RAM 4, and during regeneration operation, this target 4A R is sent to D/R via CPU 2 every control cycle.
The signal is sent to the A converter 5. A D/A converter 5 converts this signal into an analog signal, passes through a comparator 6, amplifies it with an amplifier 7, and then drives a servo valve 8.

Therefore, the hydraulic cylinder 9 is driven by the servo valve 8 in accordance with the target value R, the rotation angle of the movable element 10 and the amount of expansion and contraction of the hydraulic cylinder 9 are detected by the detector 11, and the value is detected as the regeneration trajectory Y by the comparator. 6, the hydraulic cylinder 9 reaches the target value R and regeneration II! I
The ax with the L trace Y is driven in the direction O, and the reproduced locus Y after positioning is transmitted to CI) U2.

In the present invention, the dynamic delay is changed for each regeneration operation, and the dynamic delay time at which the constancy value of a series of deviation groups is minimized is found, and this dynamic delay time is used for subsequent regeneration operations. This allows accurate positioning work to be performed.

Next, with reference to the graph of FIG. 4, the general structure of the °C real name proposal will be explained. FIG. 4 shows the tip trajectory (vertical axis) of a movable element with a certain degree of freedom corresponding to the passage of time (horizontal axis), where the thick solid line represents the taught value R6, and Y represents the playback trajectory based on this taught value R8. Not shown.

First, the operator performs teaching work and changes the teaching trajectory to the teaching value R8.
CI) U2 performs the regeneration operation using this taught value R0 as the target value, and the regeneration trajectory Y is obtained. is fetched every control cycle Lc and stored in area M9 of RAM4.

The following explanation is a data processing procedure for one degree of freedom, but in actual control, similar processing is performed for all degrees of freedom.

Now, we consider the dynamic delay time as an integer multiple of the control period Lc for convenience, and assume a delay time that is definitely faster (shorter) than the predicted dynamic delay time in the degree of freedom that is the subject of control. Let this be nto.

The rebirth trajectory of the ten books Y. and teach i1riRo (RAM
(stored in area M8 of RAM 4) at time t + ntc is set as a1 and set as IE (L +n Lc) in area L of RAM 4! Store in NM4. Then, add the deviation E(trntc) by I-recording delay time nto (backward) teaching (ifIR8(1)),
Set this as the target value R1(t) and prepare for the first time! 4:
I do. In this way, the first preliminary reproduction trajectory ¥1 as shown in the figure is obtained, and the deviation ε (tl, 1) between this first preliminary reproduction trajectory Y1 and the taught value R6 from time t1 to t''
~ε (t,,1) is detected, ,<nuC<J),
The accumulated values ε, -Σ1ε (t, 1)l are calculated by integrating their absolute values, and +J of RAM4 is calculated.
i Store in area M5.

Next, the deviation E(trnu,) from the area M4 of RAM4
are taken out sequentially at each time and set to the teaching value R8 (rl+1).
Create a target value R2 (0) in addition to tc time or earlier, and perform the second preliminary reproduction /1: according to this. Then, the trace Y2 of the second preliminary reproduction i11 as shown in the figure is obtained, Deviation ε (t,, 2) to ε (t, , 2) between this second preliminary reproduction trajectory Y2 and the taught value R0 up to time 1.
ε2−Σ1a(t,2) l is calculated, and ε1 stored in area M5 of RAM4 is calculated.
Shift the value of ε2 to area M6 -U, and at the same time shift the value of ε2 to M5
Store it in

The values of ε1 and ε2 obtained in this way are compared to determine their magnitude, but since ntc is set to a time sufficiently smaller than the expected dynamic delay time, usually ε2<ε.

In this way, the preliminary playback is repeated sequentially, and 83, εto,...
Then, the integrated value εm-1 obtained in the m-1st and m-th preliminary reproduction was compared with ε, and when it became 6m28m-1, it was adopted in the m-1st preliminary reproduction. Determine the delay time (n4m-2)Lc as the optimal delay time in the degree of freedom ti~1, and then calculate the teaching value or target obtained by the regeneration movement in consideration of the actual dynamic characteristics. A normal reproduction value is performed by adding the deviation from the value to the target value at that time by 2 (n4m-2) tc time earlier and using it as the target value to be used for the next reproduction movement.

Next, with reference to the flowchart of FIG. 3, the processing procedure when implementing the method of the present invention using a microcomputer will be explained in detail. Note that Sl, S2.・・・
・Does not represent a step number. Furthermore, as described above, this flowchart concerns one degree of freedom, and in reality, similar processing is performed for other degrees of freedom.

First, in step S1, as an initial setting, a value of n is appropriately set in order to set a delay time ntC that is reliably shorter than the predicted dynamic delay time in that degree of freedom. It also indicates the section that should be controlled.

, tr (tl < tJ) is set to an arbitrary value of 116, the control position does not indicate 6 (time), E is set to 0, and RAM 4 is set.
Set the value of area MG to 0. In robots, etc., the control accuracy decreases especially during acceleration/deceleration or at turns where the direction of movement changes. Although it is possible to obtain accurate positioning accuracy,
Furthermore, the entire work trajectory...(t□~t+), (t;+t~
'b), (t1m to t,...), and control may be performed for each section.

Now, the processing carried out in the A1 device will be described.

CI) In S2, IJ detects the state of the flag in area M7 of ζ18 M4 and determines whether the playback to be performed from now on is the first or second time. The flag in area M7 is 104 for the first time.
Only set to l by the cPU.

If this is the first re-d゛, then in S3
The re/1゛ control is performed based on the ζ teaching value sen (t+ntc) and ilT, ! 4: Locus Y. (10 r mouth.) and (↓l), and in S4, the deviation IE (t+nL
t) as E(t 4-n tc)=R, (t +n Lc)-Y
o (L + n tc ), this value is stored in the area M4 of RAM 4, and the target value RI (1) for the first preliminary reproduction is calculated.
should be calculated (taught value Ro (t) earlier by ntc
Add E(t+nt1) to 4dlf of M3,
(Stored in area 4.

Ima+ (L) -R8(t) +lE (t +n
tc) If it is not the first 100 moves in S2, that is,
For example, the n111jl L-,,I-rbii1i-1
．． 0) if m-1st time O)Yi, 4iii++ノ1
Dei! The preliminary rerun 1- based on the set target Ili'j RmJ is sent to S5 °C11, and its peak/1-trajectory Y
The bias y'ε (L,
II1) is rolled for 9, and then goes to S6.
, ) is given 9, and this is set earlier by the delay time of nto to Y (in addition to the indicated trajectory, the target value R for the next preliminary re-renewal η゛
m is created (updated) and stored in the record jQ fiji I-father M3.

Next, the CII U proceeds to processing in S7, where the current control position; and t'', and if Yes, in S8, ε (t
, m) are integrated, and if No, the process bypasses S8 and proceeds to S9.

S9 is a step of determining whether or not the final work position has been reached. If the answer is No, the value of t (M) is determined at SIO in order to move to the next control (work) position.

) is substituted with 1+1c and the process returns to S2. If Yes, t,
m) If the current integrated value is still larger than l, 1 is added to the value of n that determines the delay time ntC in step 312, assuming that there is room to further extend the l1iJI delay time. At the same time, the vector t=Q returns to the initial position, and the process returns to step S2.

In addition, if the current integrated value exceeds the previous integrated value in Sll, it means that the previous dynamic delay time setting is optimal, so n -1 is selected in S13 and the subsequent actual playback is performed. In operation, C consistently uses (n-1)to as the dynamic delay time.

The cumulative number of times t~1. If the separation is incorrect, there is an IA error that may cause an error in selecting the delay time, so if necessary, set t□ to ('i + t+)/2. (tz +t+
)/2~1. ,1. ~(1,+1k)/2. (tl
+tk)/2~t1, etc., to determine whether the selected qtlj delay time is appropriate.8.
It is desirable to do I.

As described above, the present invention causes a controlled object having multiple degrees of freedom to perform regenerative operation according to a taught value, measures the deviation between the taught value and the regenerated trajectory, and then performs a regenerative operation for a period of time corresponding to the dynamic delay time during the next regenerative operation. In a learning control method in which regeneration operation is performed by adding the above deviation to the teaching value or the current target trajectory as quickly as possible, the dynamic delay time is changed for each degree of freedom and several orders of preliminary regeneration are repeated. This learning control method is characterized by calculating the accumulated value of the deviation of ``-'' and employing the dynamic delay time at which the accumulated value of Using a microcomputer, it is possible to perform positioning control that is unaffected by noise.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing the signal transmission system of a control circuit that can be used to implement the present invention, Fig. 2 is a Jolt diagram showing the contents of the time memory device WRAM used in the control circuit, and Fig. 3 4 is a flowchart showing an example of the processing procedure of the learning control force method, which is an embodiment of the present invention, and FIG. 4 is a graph for explaining the present invention in detail. (Explanation of symbols) 1...Microcomputer 2...CPU3...
ROM 4...RAM 10...Movable elements M1-M! 1...Note 1. a area 31. 32. ...313...Step W-"R
...Target value (taught value) Y...iri/-L trajectory. One person: Kobe M Steel Co., Ltd. Patent attorney Takeo Honjo

Claims (1)

[Claims] A controlled object having multiple degrees of freedom is regenerated according to the taught value, and the deviation between the taught value and the trace of the regenerated egg is measured, and the next time the regenerated operation is performed, the time corresponding to the dynamic delay time is In a learning control method that performs regenerative operation by adding the above deviation to the taught value or current target trajectory as quickly as possible, the dynamic delay time is changed for each degree of freedom and several orders of preliminary regeneration are repeated, and a series of deviations are calculated at each preliminary regeneration. A learning control method characterized in that the integrated value of is calculated, and the dynamic delay time at which the integrated value is the minimum is adopted as the dynamic delay time during playback.